Methods for interpreting information from highly distorted two-dimensional dot matrix images
By using digital image processing and prior reasoning methods, the shortcomings of two-dimensional dot matrix images in terms of anti-fouling and anti-distortion are solved, and fast and accurate reading of printed image information is achieved, especially the uncorrected direct reading of anti-counterfeiting marks and text information.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING INSTITUTE OF GRAPHIC COMMUNICATION
- Filing Date
- 2023-01-05
- Publication Date
- 2026-06-30
AI Technical Summary
Existing methods for reading two-dimensional dot matrix images are insufficient in terms of resistance to contamination and distortion, making it difficult to quickly and accurately read highly distorted printed image information.
By combining digital image processing technology with prior reasoning methods, and through steps such as image morphological processing, binarization, center calibration and cropping, estimation of the main grid angle, calculation of differential polarity alternation sequence, recursive determination of prior region matrix, and dynamic vector parameter search, fast and accurate uncorrected reading is achieved.
It enables rapid and accurate reading of highly distorted printed images, overcoming the shortcomings of existing technologies in terms of anti-fouling and anti-distortion, and can identify and read anti-counterfeiting marks, images, or text information hidden in printed images.
Smart Images
Figure CN116050443B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of anti-counterfeiting printing, specifically a method for reading two-dimensional dot matrix image information that resists high-degree distortion. Background Technology
[0002] The method for interpreting information from highly distorted two-dimensional dot matrix images differs from methods that combine adaptive Hough transform with inverse perspective transform, as well as from time-consuming inverse surface transform methods and methods that use height search functions. Instead, it is a method based on prior knowledge and reasoning to directly interpret spatial dot matrix images captured under complex conditions without correction.
[0003] This method, when interpreting information from two-dimensional dot matrix images, requires neither image translation nor rotation, nor the establishment of complex models for inverse surface transformations. It is a simple and easy-to-implement information interpretation method that directly utilizes the original image, exhibiting strong anti-distortion capabilities. However, due to the influence of complex shooting conditions and the relative positions of the camera and the substrate, the method suffers from significant dot matrix distortion when extracting hidden information. This requires overcoming and resolving this issue through specialized, efficient, and reliable information interpretation techniques that conform to both linear and nonlinear distortion characteristics. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of two-dimensional dot matrix image information recognition technology in terms of anti-fouling, anti-distortion performance and information recognition effectiveness, and to provide a method for information recognition and reading of high-distortion printed quantum dots.
[0005] The technical solution adopted by this invention to solve its technical problem is as follows:
[0006] A method for identifying information in highly distorted two-dimensional dot matrix images, the method utilizing digital image processing technology combined with prior reasoning to quickly and accurately identify image dot matrix information, the method comprising the following steps:
[0007] (1) Image morphological processing and binarization
[0008] The raster image with wrinkles and dirt is preprocessed and binarized to become a normalized binarized image.
[0009] (2) Image centering and cropping
[0010] For the binarized image produced in step (1), center calibration is performed according to the ratio, and blank areas at the edges of the image are cropped;
[0011] (3) Estimate the angle of the main grid of the dot matrix
[0012] For the prior region of the image produced in step (2), the cumulative sum sequence is calculated pixel by pixel and processed to estimate the offset angle of the main direction of the quantum dot array;
[0013] (4) Calculation and processing of differential polarity alternation sequences
[0014] The angle of the generated dot matrix main grid is estimated in step (3). The differential polarity alternation sequence is calculated by accumulating the pixel values of the prior region and adding or deleting the sequence according to the second difference result.
[0015] (5) Recursive determination of the secondary direction matrix of the prior region
[0016] For the main direction differential polarity alternation sequence generated in step (4), if the dot matrix in the image is fully covered, step (4) is repeated to calculate the corresponding sequence of the secondary direction; otherwise, the minimum quantum dot spacing between adjacent marker lines in step (4) is used as the threshold for recursive judgment to generate a complete prior region quantum dot identification matrix.
[0017] (6) Dynamic vector parameter search and centroid correction
[0018] A recursive search is performed on the four neighborhoods of the prior region based on dynamic vectors to generate local quantum dot recognition matrices that conform to the search order.
[0019] (7) Multi-matrix transpose, expansion and splicing
[0020] Using the edge rows and columns of the recognition matrix generated in step (6) as initial conditions, we continue to search the remaining areas of the image and transpose and expand the local matrix according to the spatial distribution of quantum dots to splice it into a complete recognition matrix;
[0021] As an improvement to the above technical solution, the method also includes step (8) binary matrix information transformation: transforming the rhomboid distribution matrix generated in step (7) into a square form with single-sided sawtooth pattern that has higher actual information implantation and storage efficiency.
[0022] As an improvement to the above technical solution, the image morphology and binarization processing in step (1) includes taking a dot matrix image of wrinkles and dirt and performing local threshold binarization processing, then using a connected component function to delete regions with large areas and calculating the number of remaining connected components as the number of quantum dots.
[0023] As an improvement to the above technical solution, the image center calibration and cropping in step (2) includes integrating all pixel values row by row and column by column of the binary image and finding the horizontal and vertical coordinates corresponding to the peak values as the image center, and then cropping the original image according to the non-zero position of the image integral or the minimum length and width.
[0024] As an improvement to the above technical solution, the estimation of the main grid angle of the quantum dot matrix in step (3) includes using pulse width modulation combined with a minimum duty cycle optimization algorithm to process the central region of the image to estimate the main direction offset angle of the quantum dot matrix.
[0025] As an improvement to the above technical solution, the differential polarity alternation sequence calculation and processing in step (4) is to use a group of straight lines with the same slope to calculate the sum of pixel values, then calculate the pixel value sequence at the differential polarity alternation moment, and add or delete elements according to the difference between adjacent elements in the sequence.
[0026] As an improvement to the above technical solution, the recursive determination of the secondary direction matrix of the prior region in step (5) means that if the quantum dot array is implanted in a fully paved manner or the number of points it contains meets the requirements of subsequent decoding, then the method described in steps (3) and (4) is repeated to calculate and process the eigenvalue sequence of the secondary direction; otherwise, the recursive determination is performed using the minimum quantum dot spacing between adjacent marker lines in step (4) as the threshold.
[0027] As an improvement to the above technical solution, the dynamic vector parameter search and centroid correction in step (6) refers to the recursive search of the four neighborhoods centered on the prior region, with reference to the dynamic vector. If a quantum dot is found, the initial position of the next search will be shifted to the centroid position of the quantum dot.
[0028] As an improvement to the above technical solution, the multi-matrix transpose, expansion and splicing in step (7) refers to using the edge rows and columns of the recognition matrix generated in step (6) as the initial conditions, continuing to search the remaining area of the image and transposing and expanding the local matrix according to the spatial distribution of quantum dots to splice it into a complete recognition matrix.
[0029] The beneficial effects of this invention are as follows:
[0030] This invention is a method for reading information from two-dimensional dot matrix images that resists high-degree distortion. It is a non-corrected direct reading method for quickly and accurately reading anti-counterfeiting marks, images, or text information hidden in printed images. This reading method overcomes the shortcomings of existing printed quantum dot technology in terms of resistance to staining and distortion, as well as the deficiencies in the effectiveness of information reading methods. Attached Figure Description
[0031] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0032] Appendix Figure 1 This is a flowchart of the first embodiment of the present invention;
[0033] Appendix Figure 2 This is a flowchart of the dynamic vector search method of the present invention. Detailed Implementation
[0034] Example 1
[0035] A method for reading information from distorted dot matrix images, primarily applied to vertically implanted dots at equal intervals in the primary and secondary directions.
[0036] See attached document Figure 1The method includes the following steps:
[0037] (1) Image morphological processing and binarization
[0038] The distorted images obtained from the shooting are subjected to local average threshold binarization and morphological processing such as opening and closing operations, and connected component counting and large-area connected component removal are performed.
[0039] (2) Image centering and cropping
[0040] The quantum dot array binary image produced in step (1) is subjected to image noise reduction processing, and then cropped according to the minimum length and width and then expanded to a certain extent.
[0041] (3) Estimate the principal direction grid angle of the prior region of the lattice.
[0042] For the binary image with a centered dot matrix and blank edges produced in step (2), the prior region of the image center is dynamically selected. When the degree of image distortion is small, the rough angle is estimated by the binary method and then the precise angle is determined by the greedy algorithm. Otherwise, the search range is narrowed down step by step to traverse the image. The grid angle is due to the geometric distortion position difference caused by the offset of the camera and the substrate and the shooting posture during shooting.
[0043] (4) Calculation and processing of differential polarity alternation sequences
[0044] For the main direction grid angle generated in step (3), the pixel accumulation sequence is calculated with the direction and its corresponding prior region. Then, the difference sequence of the sequence is calculated and the pixel position sequence when the positive and negative values alternate is determined. The equal arithmetic adjustment within the error tolerance is performed according to the second difference result.
[0045] (5) Recursive determination of the secondary direction matrix of the prior region
[0046] For the pixel position sequence generated in step (4), the minimum distance within the range of obtuse angle difference between the primary and secondary directions is calculated by using the quantum dots through which the two calipers with adjacent elements as intercepts pass in the sequence in the central region of the image. The search and recursion are performed within its error tolerance to determine the prior region of the secondary direction.
[0047] (6) Vectorized parameter search and dynamic correction
[0048] For the prior region determination matrix generated in step (5), the cumulative vectors in the horizontal and vertical directions are dynamically replaced in the quantum dot recursive search process in combination with the zero-avoidance principle. At the same time, when the existence of the quantum dot is determined, the initial position of the next search is shifted to the centroid position of the quantum dot.
[0049] (7) Multi-matrix transpose, expansion and splicing
[0050] For the four-neighbor recognition matrix of the central region generated in step (6), the remaining areas of the image are further searched and judged according to the dynamic vector and edge search results. The local matrix is transposed, expanded and spliced into a complete recognition matrix according to the actual spatial distribution of the quantum dot matrix.
[0051] Example 2
[0052] A method for reading information from distorted dot matrix images, primarily applied to images with equal spacing in the main direction and honeycomb-like implantation in the secondary direction.
[0053] The steps (1) to (7) of this method are the same as in Example 1. Step (8) Binary matrix information conversion: convert the rhomboid distribution matrix information directly read without correction into square matrix information with single-sided sawtooth distribution.
[0054] Example 3
[0055] A method for reading distorted dot matrix images of honeycomb-embedded product labels that are somewhat soiled.
[0056] The method steps (1) to (2) are the same as in Example 1. Step (3) Estimating the grid angle of the primary direction of the quantum dot array prior region: If the center of the quantum dot array is heavily contaminated, resulting in an incorrect grid angle estimation, the image is taken again. Steps (4) to (5) are the same as in Example 1. Step (6) Dynamic vector parameter search and centroid correction: If the centroid correction value exceeds a certain multiple of the average radius of the quantum dots, the correction is no longer performed. Step (7) is the same as in Example 1. Step (8) Binary matrix information conversion: The rhomboid distribution dot array information read in step (7) is converted into square dot array information with a single-sided sawtooth distribution.
[0057] Example 4
[0058] A method for reading distorted dot matrix images, primarily applied to small-area, low-capacity commodity tags implanted in a honeycomb pattern.
[0059] The method steps (1) are the same as in Example 1. Step (2) Image center calibration and cropping: Calculate the non-zero positions in the horizontal and vertical directions of the image in positive and negative order, and then crop the image according to the position. Step (8) Binary matrix information conversion: Convert the rhomboid distribution matrix information read in step (7) into square matrix information with single-sided sawtooth distribution.
[0060] Example 5
[0061] A dynamic vector search determination method.
[0062] See attached document Figure 2 The method includes a dynamic vector recursion procedure based on zero-connection avoidance and centroid correction, specifically including the following steps:
[0063] (1) Generation of initial coordinates and their determination matrix;
[0064] (2) Avoiding consecutive zeros;
[0065] (3) Centroid correction;
[0066] (4) Dynamic vector recursion;
[0067] (5) Matrix transpose, expansion and splicing.
[0068] Among them, the initial coordinates and the judgment matrix mentioned in step (1) are two rows / columns of the prior region edge close to the region to be identified; the zero avoidance and centroid correction mentioned in step (2) refer to taking the recursive direction away from the identified region as the main direction. Assuming that the two judgment results before the current position do not exist, the quantum points in the row and column to which the current position belongs have been recursively judged as the basis, and continue to recursively along the secondary direction; the dynamic vector recursion in step (4) includes using the difference in centroid coordinates of continuous quantum points in the search direction to replace the existing vector to achieve dynamic update.
[0069] While various embodiments of the present invention have been described above, it will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above. The above description is merely a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made within the scope of the claims of this invention should still fall within the patent coverage of this invention.
[0070] It should also be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.
Claims
1. A method for reading information from highly distorted two-dimensional dot matrix images, characterized in that: The method utilizes digital image processing technology combined with prior inference algorithms to quickly and accurately identify image dot matrix information. The method includes the following steps: (1) Image morphological processing and binarization The raster image with wrinkles and dirt is preprocessed and binarized to become a normalized binarized image. (2) Image center calibration and cropping The binarized image generated in step (1) is centered according to the proportion, and the blank areas at the edge of the image are cropped to generate the prior region of the image. (3) Estimate the principal direction grid angle of the prior region of the lattice. For the prior region of the image generated in step (2), the cumulative sum sequence is calculated pixel by pixel and processed to estimate the grid angle of the principal direction of the prior region of the quantum dot array; (4) Calculation and processing of differential polarity alternation sequences For the quantum dot array prior region main direction grid angle estimated in step (3), the differential polarity alternation sequence is calculated by accumulating the pixel values of the prior region, and the differential polarity alternation sequence is added or deleted according to the second difference result; the calculation and processing of the differential polarity alternation sequence refers to calculating the pixel value accumulation sequence based on the main direction grid angle and its corresponding prior region, then calculating the difference sequence of the pixel value accumulation sequence and determining the pixel value sequence when positive and negative alternation occurs, and performing equal arithmetic adjustment within the error tolerance according to the second difference result; (5) Recursive determination of the secondary direction matrix of the prior region For the main direction differential polarity alternation sequence generated in step (4), if the dot matrix in the image is fully covered, step (4) is repeated to calculate the corresponding sequence of the secondary direction; otherwise, the minimum quantum dot spacing between adjacent marker lines in step (4) is used as the threshold for recursive judgment to generate a complete prior region quantum dot identification matrix. The minimum quantum dot spacing is used as a threshold for recursive determination, which means using the centroid of the quantum dots through which the adjacent tracing lines in the main direction of the image center region pass to calculate the minimum distance within the obtuse angle range, and searching and recursively determining the prior region of the secondary direction within its error tolerance; wherein, the quantum dots through which the two tracing lines with adjacent elements as intercepts in the differential polarity alternation sequence generated in step (4) of the image center region are used to calculate the minimum distance within the obtuse angle range of the main and secondary direction angle difference; (6) Dynamic vector parameter search and centroid correction A recursive search is performed on the four neighborhoods of the prior region based on dynamic vectors to generate local quantum dot recognition matrices that conform to the search order. (7) Multi-matrix transpose, expansion, and splicing Using the edge rows and columns of the recognition matrix generated in step (6) as initial conditions, we continue to search the remaining areas of the image and transpose and expand the local matrix according to the spatial distribution of quantum dots to splice it into a complete recognition matrix; (8) Binary matrix information transformation.
2. The method for reading information from highly distorted two-dimensional dot matrix images according to claim 1, characterized in that: The image morphological processing and binarization techniques described in step (1) include local average threshold binarization of the image, erosion and dilation morphological processing, and connected component counting and large-area removal.
3. The method for reading information from highly distorted two-dimensional dot matrix images according to claim 1 or 2, characterized in that: The image center calibration and cropping in step (2) includes noise avoidance, calculation of non-zero positions in the horizontal and vertical directions, and cropping of non-full-area images according to non-zero positions, while full-area images need to be cropped according to the minimum length and width and then expanded to a certain extent.
4. The method for reading information from highly distorted two-dimensional dot matrix images according to claim 3, characterized in that: The estimation of the main direction grid angle of the prior region of the dot matrix in step (3) includes the dynamic selection of the prior region of the image center. When the degree of image distortion is small, the rough angle is estimated first by the binary method and then the precise angle is determined by the greedy algorithm. Otherwise, the search range is narrowed down step by step to perform traversal.
5. The method for reading information from highly distorted two-dimensional dot matrix images according to claim 4, characterized in that: The grid angle mentioned in step (3) is due to the geometric distortion position difference caused by the offset between the camera and the printing substrate and the shooting posture during shooting.
6. The method for reading information from highly distorted two-dimensional dot matrix images according to claim 1, characterized in that: The dynamic vector parameter search and centroid correction mentioned in step (6) refers to dynamically replacing the accumulated vectors in the horizontal and vertical directions in combination with the zero-avoidance principle during the quantum dot recursive search process, and shifting the initial position of the next search to the centroid position of the quantum dot when the existence of the quantum dot is determined.
7. The method for reading information from highly distorted two-dimensional dot matrix images according to claim 1, characterized in that: The binary matrix information conversion mentioned in step (8) refers to converting the rhomboid distribution matrix information directly read without correction into square matrix information with a single-sided sawtooth distribution.